Key Takeaways
- Daily spending through the Link Agent Wallet rose nearly 80-fold from early September to mid-November 2026, but the underlying transaction volume remains undisclosed.
- Airfare, travel, and lodging generated more than 40% of recent spending, while over half of buyers returned for another agentic purchase.
- Merchants preparing for AI-directed commerce face practical questions involving authorization, product data, refunds, fraud controls, and accountability.
Stripe has provided an early look at consumers allowing AI agents to execute purchases, rather than simply recommend products. On a seven-day rolling average, daily spending through Link Agent Wallet was nearly 80 times higher in mid-November 2026 than at the beginning of September. Spending also doubled during the preceding month.
While this represents an eye-catching growth rate, it is not a measure of the overall agentic-commerce market. Stripe did not disclose the absolute dollar value, transaction count, or number of consumers behind the increase. The data covers Link Agent Wallet activity, not every purchase initiated by an AI agent. A nearly 80-fold jump from a limited starting point can demonstrate momentum without establishing that delegated payments have reached mass adoption.
Distribution also expanded during the measurement period. According to Dealroom News, relevant services associated with these agent-directed purchases include Meta’s Muse, xAI’s Grok, and Instinct. More supported agents create more opportunities for spending, so the increase likely reflects a mixture of new integrations and heavier use by existing customers.
Where are agents finding an early role? Travel stands out.
More than 40% of recent Link agent spending went toward airfare, travel, and lodging. Retail represented roughly one-third, with other categories accounting for about one-quarter. Travel serves as a plausible testing ground because booking decisions often require customers to compare dates, airports, fares, hotel locations, baggage costs, loyalty benefits, and cancellation terms at the same time.
An agent can filter those variables rapidly. A traveler might authorize it to find a refundable hotel within a fixed distance of a venue and below a specified price, then complete the reservation using controlled payment credentials. Retail offers similar use cases involving size, compatibility, delivery deadlines, brand preferences, and spending limits.
Finding an acceptable product is not the same as granting software permission to buy it. That final step introduces financial and operational consequences that a mere recommendation does not.
The repeat-use data therefore carries particular weight. More than half of consumers who completed an agentic purchase returned for a second purchase, generally within the same category. This pattern suggests trust may be forming in narrow lanes. A customer who delegates one hotel booking may delegate another without being comfortable authorizing unrelated purchases across several merchants.
Broader consumer sentiment remains cautious. A Wildfire Systems survey of 1,007 U.S. consumers found that only 5% would trust an AI platform to process an agent’s payment, while 41% trusted none of the payment options tested. That gap between early repeat users and the wider population could shape how quickly the model expands.
For payment providers, controlled-spend wallets and virtual cards can help limit exposure. An agent might receive credentials restricted to one merchant, category, transaction, or spending ceiling. Protocols such as x402 offer another model, enabling machine-to-machine payments on a per-request basis. These mechanisms can reduce open-ended authority, although they do not resolve every dispute.
What happens if an agent books the wrong airport, overlooks a restrictive fare, or orders an incompatible component? Merchants will need records showing what the customer authorized, what information the agent received, and why the transaction was approved. Refund handling and chargebacks may become especially complicated when several parties participate in the decision.
Preparation goes beyond payments. Merchants can improve agent access by publishing structured, current information about inventory, prices, delivery dates, cancellation terms, product specifications, and return policies. They also need ways to distinguish authorized purchasing agents from scraping bots or fraudulent automation.
The latest figures show that agent-directed purchasing is producing real activity, especially in travel and retail. The next phase will depend less on spectacular percentage growth and more on whether merchants, payment networks, and agent developers can make delegated transactions limited, traceable, and reasonably easy to reverse when something goes wrong.
⬇️